The use of digital signal processing in hearing aids and other devices has become commonplace. One goal of such systems is to provide amplification of desirable audio information in a signal while suppressing undesirable audio noise in the signal.
A person using a hearing aid or other audio device will typically be in an environment with several different types of real-life audio signals consisting of noises and desirable sounds. Examples of such audio signals are: stationary noise (such as a fan or motor), pseudo-stationary noise (such as traffic noise or speech babble), desirable sounds (such as speech or music) and transient noise (such as gun shots or a door slamming).
Various methods of detecting noise have been proposed and implemented.
In one system, described in U.S. Pat. No. 4,852,175, the incoming audio signal is divided into a set of frequency bands and the “sound events” in each band are categorized by their amplitude (or intensity). An assumption is made that a pre-selected percentage of the sound events with the lowest amplitude are noise events and a gain is calculated separately for each band to attempt to minimize the effect of the identified noise on an output signal, which is formed by recombining the signal from each frequency band after having multiplied it by the calculated gain. This system is deficient because it makes a presumption that a certain percentage of sound events in each frequency band are noise based only on their amplitude. This presumption is not a reliable measure of noise in most circumstances. Furthermore, this system cannot adapt to changing conditions in which noise is more or less prevalent at different times. The result is that many noise sound events will not be categorized as noise and many non-noise sound events will be categorized as noise.
In another system, described in U.S. Pat. No. 4,185,168, the absolute value, or a function thereof (e.g. the RMS value), of the signal in each frequency band is used to estimate the noise content in the frequency band, assuming that the noise has a fixed or narrow frequency spectrum over a selected time period. Alternatively, a smoothed version of the signal in each band can be used to produce the signal-to-noise ratio, SNR, which can be used to determine the presence of noise. If noise is detected, the gain of the band relative to other bands is reduced so that bands with noise are suppressed in favor of bands without noise. While this system does not assume that a selected amount of audio information in each band will be noise, it is deficient because it assumes that noise has a frequency spectrum which does not vary with time or varies only within a narrow range over a period of time. This system is accordingly limited to detecting stationary or slowly changing pseudo-stationary noise.
There is a need for a signal characterization and noise reduction system that is adaptable to signals which have different noise content over time and which is capable of detecting and suppressing different types of noise.